Quantitative Strategies on High Frequency Data

Submission of research project

Author

Sujith Madesh, Kacper Mosoń

Published

Jan 20, 2026

Group 1

Approaches undertaken

We implemented and evaluated three intraday strategies on 1‑minute data, aggregated to daily PnL and summarized quarter-by-quarter:

  • Strategy 1 (EMA crossover): momentum/mean-reversion variants based on a fast vs slow exponential moving average.
  • Strategy 2 (Volatility breakout 2.2): fast EMA as the signal with volatility bands defined by slow EMA ± \(m\cdot\sigma\) (rolling standard deviation). Both momentum and mean‑reversion variants were evaluated.
  • Strategy 3 (RF‑EMA crossover 2.3): EMA crossover with a regime filter (trade only when the EMA spread is large relative to recent volatility).

Assumptions / execution rules

  • We avoid the most illiquid minutes by excluding the first/last 10 minutes of the session from calculations.
  • We do not trade in the first part of the session and we flatten before the close (no overnight risk).
  • PnL is computed in USD using contract point values and includes fixed transaction costs per trade.

Parameter search & selection

  • For each quarter and each contract we ran a grid search over strategy parameters.
  • We computed performance metrics (gross/net Sharpe, Calmar, cumulative PnL, turnover) and a ranking statistic (stat).
  • To avoid overfitting within a quarter, we selected one fixed parameter set per contract by maximizing the sum of stat across in‑sample quarters.

Finally selected strategy

Based on robustness across quarters and net performance, our final choice for Group 1 is:

  • Strategy 2: Volatility Breakout 2.2 (EMA signal + volatility bands)
    • NQ: MOM variant with (signalEMA=20, slowEMA=60, volat_sd=30, m=2.0)
    • SP: MR variant with (signalEMA=90, slowEMA=180, volat_sd=60, m=1.0)

These parameters are held fixed across quarters for each contract.

Summary of results

quarter sym gross_SR net_SR gross_PnL net_PnL gross_CR net_CR av_daily_ntrans stat
2023_Q1 NQ 6.11 5.85 59253.00 56901.00 37.29 35.26 3.02 21.62
2023_Q1 SP -0.98 -1.34 -6326.15 -8606.15 -1.68 -2.15 2.92 -3.96
2023_Q3 NQ 0.83 0.52 6365.52 4013.52 1.92 1.15 3.02 0.03
2023_Q3 SP 2.23 1.74 9282.50 7194.50 7.70 5.38 2.68 2.45
2023_Q4 NQ 2.05 1.78 16052.96 13988.96 8.87 6.98 2.69 3.38
2023_Q4 SP -0.08 -0.50 -340.20 -2188.20 -0.12 -0.71 2.41 -0.78
2024_Q2 NQ -0.24 -0.43 -2615.56 -4607.56 -0.50 -0.85 2.55 -1.42
2024_Q2 SP 3.07 2.70 15334.60 13366.60 9.56 7.99 2.52 5.70
2024_Q4 NQ 2.87 2.65 26062.22 24166.22 20.90 19.00 2.43 6.86
2024_Q4 SP 0.51 0.16 2757.60 861.60 0.97 0.27 2.43 -0.00
2025_Q1 NQ 0.54 0.33 5660.44 3404.44 0.98 0.57 2.98 -0.21
2025_Q1 SP 4.06 3.81 30609.65 28617.65 28.57 25.70 2.63 11.12
2025_Q2 NQ 0.33 0.25 7438.44 5518.44 0.89 0.64 2.46 -0.43
2025_Q2 SP 3.87 3.71 42551.35 40607.35 25.84 24.20 2.49 11.87

Comment: The table reports quarter-by-quarter metrics for NQ and SP only. Results are regime dependent: NQ (MOM) benefits more in trending quarters, while SP (MR) performs best when mean reversion dominates; net results are consistently below gross due to transaction costs.

Equity lines

2023Q1

NQ Comment: Strong trending quarter; the momentum breakout delivers a steep upward net equity path after costs.

SP Comment: Weak/negative quarter; mean-reversion has low edge here and costs amplify the drawdown.

2023Q3

NQ Comment: Flatter quarter with smaller gains; costs matter more when momentum edge is weaker.

SP Comment: Positive quarter; mean-reversion performs better in the more two-sided regime.

2023Q4

NQ Comment: Positive quarter; momentum breakout captures persistent moves and keeps net equity rising.

SP Comment: Slightly negative; mean-reversion struggles to overcome costs in this regime.

2024Q2

NQ Comment: Challenging quarter for momentum; choppy/mean-reverting conditions lead to drawdowns.

SP Comment: Strong quarter; mean-reversion benefits from oscillating prices and delivers solid net gains.

2024Q4

NQ Comment: Strong quarter; sustained moves make the momentum breakout highly profitable after costs.

SP Comment: Near-flat quarter; small gross gains are largely eaten by transaction costs.

2025Q1

NQ Comment: Modest positive quarter; momentum edge is present but comparatively small.

SP Comment: Very strong quarter; mean-reversion is the main driver of performance.

2025Q2

NQ Comment: Positive but lower-edge quarter; net equity rises gradually and remains sensitive to costs.

SP Comment: Strong quarter; mean-reversion produces large net gains with manageable turnover.

Group 2

Approaches undertaken

We implemented and evaluated two primary logic types on 5-minute data, designed to handle high transaction costs ($15/$10) relative to volatility:

  • Strategy 1 (Volatility Breakout): A trend-following approach where entries are triggered when the close price breaches the max/min of the last \(N\) bars, conditioned on a minimum volatility threshold (rolling std dev) to ensure the move justifies the cost.
  • Strategy 2 (EMA + RSI): A trend-momentum strategy using EMA crossovers for direction, filtered by RSI levels to prevent entering overextended positions (overbought/oversold).

Assumptions / execution rules

  • Timezone Management: Data was converted from UTC to CET to strictly adhere to the 16:50 CET forced exit rule.
  • Forced Exits: All positions are flattened daily at 16:50 CET. No trading occurs during the break (17:00–18:00) or the first 10 minutes after re-open (18:00–18:10).
  • Session Filtering: To combat noise, we introduced a regime filter testing “Full Day” trading vs. “US Session Only” (13:00–16:30 CET) to limit entries to peak liquidity hours.

Parameter search & selection

  • We performed a global portfolio optimization (cross-product grid search) rather than optimizing assets in isolation.
  • We tested lookback windows ranging from 2 hours to 24 hours (24–288 bars) and various volatility thresholds.
  • The selection criterion was maximizing the sum of the ranking statistic across all 7 in-sample quarters combined, ensuring the selected parameters were robust across different market years (2023–2025).

Finally selected strategy

Based on the optimization results, our final choice is a hybrid Volatility Breakout portfolio that treats the two metals differently based on their noise profiles:

Strategy 1: Volatility Breakout

  • XAU (Gold): Long-Term Trend variant
    • Parameters: (window=288 [24h], vol_threshold=0.5, session=Full)
  • XAG (Silver): Session-Filtered variant
    • Parameters: (window=144 [12h], vol_threshold=0.05, session=US_Only)

Summary of results

quarter sym gross_SR net_SR gross_PnL net_PnL gross_CR net_CR av_daily_ntrans stat
2023_Q1 XAG 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.00
2023_Q1 XAU 0.03 -0.25 212.50 -1947.50 0.05 -0.42 1.62 -0.50
2023_Q3 XAG 1.68 1.68 2225.00 2205.00 0.00 0.00 0.02 0.94
2023_Q3 XAU 1.67 1.14 5167.60 3547.60 5.01 3.11 1.20 0.81
2023_Q4 XAG 2.38 2.38 1405.00 1365.00 0.00 0.00 0.04 0.59
2023_Q4 XAU 1.90 1.56 10947.90 8997.90 7.01 5.64 1.44 2.33
2024_Q2 XAG 1.97 1.89 6355.00 6095.00 5.49 5.18 0.28 2.51
2024_Q2 XAU -0.19 -0.44 -1788.00 -4158.00 -0.35 -0.79 1.72 -1.33
2024_Q4 XAG -1.03 -1.11 -1740.00 -1880.00 -1.98 -2.07 0.15 -1.01
2024_Q4 XAU 3.42 3.17 29483.30 27443.30 13.45 12.03 1.48 8.85
2025_Q1 XAG -0.11 -0.16 -90.00 -130.00 -0.37 -0.53 0.04 -0.00
2025_Q1 XAU 2.61 2.36 22215.70 20100.70 10.78 9.32 1.55 5.58
2025_Q2 XAG -0.72 -0.86 -580.00 -700.00 -2.66 -3.07 0.13 -0.00
2025_Q2 XAU 0.58 0.46 11077.10 8782.10 2.00 1.50 1.66 -0.09

Comment: Gold (XAU) generally exhibited strong positive performance in trending quarters, notably 2024 Q4, driven by its 24-hour breakout strategy. Silver (XAG), with its US-session-only filter, largely avoided significant losses, but saw mixed PnL due to the highly restrictive time filter which severely limited the number of trades.

Equity lines

2023Q1

XAG (Silver): Remained completely flat (zero trades) as the strict US-session filter prevented entry during the lack of clear direction.

XAU (Gold): Suffered a deep “V-shaped” drawdown in February due to sharp reversals, though a strong rally in March helped recover a significant portion of the losses.

2023Q3

XAG (Silver): Stayed out of the market for nearly the entire quarter before catching a single, highly profitable breakout trend in late September.

XAU (Gold): Traded actively with mixed results, enduring a mid-quarter dip before finishing with a moderate profit as volatility increased

2023Q4

XAG (Silver): Executed a few precise, profitable trades, resulting in a clean “step-like” equity curve with minimal drawdown.

XAU (Gold): Capitalized on a sustained 3-month uptrend, generating consistent profits with the 24-hour breakout logic working effectively.

2024Q2

XAG (Silver): Displayed high volatility, recovering from a sharp mid-quarter drawdown to finish with a strong profit, effectively decoupling from Gold’s poor performance.

XAU (Gold): Experienced its worst quarter, caught in a deep whipsaw pattern during May that resulted in a steady accumulation of losses.

2024Q4

XAG (Silver): Struggled to find traction, taking an early loss and failing to recover, ending the period with a small net loss.

XAU (Gold): Delivered exceptional performance, riding a smooth, continuous trend to generate the highest profit of any quarter (over $25k).

2025Q1

XAG (Silver): Took a single sharp loss in February and remained flat thereafter, effectively sidelined by the session filter.

XAU (Gold): Continued its winning streak with a nearly vertical equity curve, demonstrating the strategy’s ability to compound gains during strong bull markets.

2025Q2

XAG (Silver): Remained choppy and negative, unable to sustain a trend during the US session window.

XAU (Gold): Peaked early with massive gains ($30k+) but gave back a significant portion during a late-quarter reversal, though still finishing profitably.

Summary and conclusions

Group 1 (NQ + SP)

  • We select Volatility Breakout 2.2 with fixed parameters: NQ uses MOM, SP uses MR.
  • Performance is clearly regime dependent: NQ MOM contributes most in trending quarters, while SP MR contributes most in choppy/mean‑reverting quarters.
  • Transaction costs matter: the gap between gross and net is material, especially in lower-edge quarters.
  • The combined design (MOM on NQ + MR on SP) provides a more balanced profile than running a single style on both assets.
  • Overall, we view this as a reasonable, interpretable intraday strategy for Group 1, but risk control and cost assumptions are critical for real deployment.

Group 2 (XAU + XAG)

  • Final Selection: We select a Hybrid Volatility Breakout strategy with asset-specific time regimes: Gold (XAU) operates on a continuous 24-hour window, while Silver (XAG) is strictly limited to the US-session (13:00–16:30 CET).

  • Role Allocation: Performance is driven by distinct roles: XAU captures major global trends and overnight moves, while XAG remains flat during noise and only deploying capital during peak liquidity.

  • Cost Constraint: Transaction cost sensitivity (15/10 fixed costs) is the defining challenge for this group. Standard high-frequency signals failed; success required extending lookback windows (12h–24h) to ensure average trade size significantly exceeded the break-even threshold.

  • Portfolio Synergy: The combined design offers diversification through execution logic rather than just asset class. By filtering Silver’s entries to the US session, we reduced portfolio correlation during low-volatility Asian/European mornings.

  • Overall View: We view this as a robust, regime-aware strategy. While it relies heavily on Gold for gross profit, the addition of the time-filtered Silver component improves the quality-per-trade and prevents capital erosion during non-trending periods.